FORMA (Framework for Observation of Recursive Model Architectures) proposes that semantic failure modes in coupled interpretive systems — including but not limited to large language models — map structurally to René Thom's elementary catastrophes. The framework operates at the level of observable coupling dynamics, making no claims about consciousness, phenomenology, or interiority. This upload contains the canonical FORMA theory document (v1.2.5), the frozen theory baseline (v1.1.4) documenting that core claims have remained unchanged through all subsequent patches, Extension I (the Subclinical Continuum), and external convergence evidence documenting that multiple independent 2022-2026 research programs have established singularity theory and phase transition mathematics as governing LLM dynamics — without yet bridging to behavioral failure mode classification at inference time, which is FORMA's specific contribution. The pre-Thom phenomenology (escape classes) has achieved 100% predictive accuracy in Phase 1 validation across 10+ architectures. The catastrophe-theoretic mapping is under active adversarial testing. Known concerns (zombie formalism, co-development bias) are documented in the canonical text. The author's credence on the catastrophe claims is lower than that of the AI systems used to stress-test them.